ch.01 · ai-labs
The Lossless Group's pseudomonorepo for AI experiments graduating into real products. Currently housing 8 active builds; earlier strands have already shipped elsewhere.
ch.02 · active builds
8 children. Each on its own three-tier branches; each rolls its changelog/ and context-v/ up to this surface.
The installable practice. The `cv` Claude Code plugin gives your agent the context-vigilance skill, `/cv:*` workflow commands that each put it in a role, templates, and a starter scaffold. No database, no config, no server. v0.2.0 is the first tagged release and is in review for Claude's plugin directory. The tree-wide corpus it grew out of now lives separately in context-v-corpus, with its public slice searchable online.
Code-first slide-deck operating system for receiving-side Due-Diligence-grade content. A shared Scroll-UI deck shell, a canonical library of 27 slide types, PPTX and Keynote export, and `/remake-deck`, which turns a source PDF or PPTX into three fresh design variants with one command. Client engagements drive the pattern from nested submodules and stay out of this public rollup.
Multi-tenant AI data augmentation. Upload a CSV, fire enrichment passes against it (custom prompts or source packs like LinkedIn, X, Bluesky, Wikipedia), triage the responses, and promote enhanced rows into new canonical record sets. Federated Svelte 5 microfrontends on Rsbuild, over stateless TypeScript services talking NATS. Every column on every row is tenant-defined; there is no hardcoded schema.
The identity plane behind `id.didi.sh`. One small owned auth service: create an account once from inside whichever app invited you, and you're signed in across memos (MemoPop), decks (DidiDecks), and Augment It via a single `.didi.sh` session cookie. The tree's deliberate polyglot exception — Elixir/Phoenix where everything else is TypeScript — because a session service is exactly the workload BEAM was built for.
A document that keeps its workings. Publish a `.flave` and people see the conclusion; send the file itself and they get the evidence, the data and its sources, the reasoning, and the design vocabulary that produced it. Content, data, and assets each carry a clearance (private, team, LP, public), so one document yields every version without drift. Built on Lossless Flavored Markdown; the Tauri desktop app opens a folder and restyles the document live as you edit its theme.
Build corpora deliberately: capture, triage, fetch, file, and quality-check the source material that grounds decks, memos, and enriched record sets. Corpus building had been rebuilt inside three products before it got a repo of its own. Plain markdown files are the source of truth, PDFs are content-addressed and mirrored to R2, and a change feed tells a client what changed in their corpus and why.
An open toolkit for the 100-day Personal Strategic Plan used in ChoiceCenter's Leadership Legacy program. Not an app: give Claude the repo, or install the skill, and it helps you write goals you can keep, score each week on what you actually did, and start again the next morning.
ch.03 · graduated
Strands that started in this tree and now live elsewhere.
ch.04 · discipline
Every child carries the same context-v/ shape. The parent owns cross-cutting documents that span more than one child.
Structural changes — submodule moves, monorepo corrections, new children — land as parent-level entries. Feature ships live in each child's own changelog.
Three tiers at parity per the pseudomonorepos discipline. branch = on every .gitmodules entry. Parent gitlinks tidied deliberately, not auto-bumped.
ch.05 · studies
Each study scopes one domain question and pins upstream code as nested submodules. We read from the pinned source, not from training data.
Which file conventions have converged for human + agent cooperation?
AGENTS.md · SKILL.md · llms.txt · MCP · A2A · OpenSpec · Spec Kit
How do production memory systems decide what to remember, when to update, and at what scope?
Mem0 · neo · StateBench
Which machine-validatable specs converge across the analytics stack?
Data Package · Parquet · Arrow · Vega-Lite · Plot · ggsql